Spanner is Google's globally distributed database that provides SQL queries and ACID transactions across datacenters. It uses TrueTime to assign timestamps based on a global clock with bounded uncertainty. This allows lock-free read transactions to retrieve a consistent view of data. Write transactions use two-phase locking and commit in timestamp order respecting the global clock to ensure external consistency.

1. Spanner: Google’s
Globally-Distributed Database
Wilson Hsieh
representing a host of authors
OSDI 2012

2. What is Spanner?
• Distributed multiversion database
• General-purpose transactions (ACID)
• SQL query language
• Schematized tables
• Semi-relational data model
• Running in production
• Storage for Google’s ad data
• Replaced a sharded MySQL database
OSDI 2012 2

3. Example: Social Network
OSDI 2012
User posts
Friend lists
US
Brazil
Russia
San Francisco
Seattle
Arizona
Spain
Sao Paulo
Santiago
Buenos Aires
Moscow
Berlin
Krakow
London
Paris
Berlin
Madrid
Lisbon
3
x1000
x1000
x1000
x1000

4. Overview
• Feature: Lock-free distributed read transactions
• Property: External consistency of distributed
transactions
– First system at global scale
• Implementation: Integration of concurrency
control, replication, and 2PC
– Correctness and performance
• Enabling technology: TrueTime
– Interval-based global time
OSDI 2012 4

5. Read Transactions
• Generate a page of friends’ recent posts
– Consistent view of friend list and their posts
OSDI 2012
Why consistency matters
1. Remove untrustworthy person X as friend
2. Post P: “My government is repressive…”
5

6. Single Machine
User posts
Friend lists
Friend2 post
Generate my page
Friend1 post
Friend999 post
Friend1000 post
Block writes
OSDI 2012
…
6

7. Multiple Machines
User posts
Friend lists
User posts
Friend lists
Generate my page
Block writes
Friend1 post
Friend2 post
…
Friend999 post
Friend1000 post
OSDI 2012
7

8. Multiple Datacenters
User posts
Friend lists
User posts
x1000
Friend lists
User posts
Friend lists
User posts
Friend lists
Generate my page
Friend1 post
US
Friend2 post
Spain
Friend999 post
Brazil
Friend1000 post
OSDI 2012
…
Russia
8
x1000
x1000
x1000

9. Version Management
• Transactions that write use strict 2PL
– Each transaction T is assigned a timestamp s
– Data written by T is timestamped with s
Time <8 8
[X]
[me]
15
[P]
My friends
My posts
X’s friends
[]
[]
OSDI 2012 9

10. Synchronizing Snapshots
Global wall-clock time
==
External Consistency:
Commit order respects global wall-time order
==
Timestamp order respects global wall-time order
given
timestamp order == commit order
OSDI 2012 10

11. Timestamps, Global Clock
• Strict two-phase locking for write transactions
• Assign timestamp while locks are held
T
Acquired locks Release locks
Pick s = now()
OSDI 2012 11

12. Timestamp Invariants
• Timestamp order == commit order
T2
T1
• Timestamp order respects global wall-time order
T3
T4
OSDI 2012 12

13. TrueTime
• “Global wall-clock time” with bounded
uncertainty
time
TT.now()
earliest latest
2*ε
OSDI 2012 13

14. Timestamps and TrueTime
T
Acquired locks Release locks
Pick s = TT.now().latest
s Wait until TT.now().earliest > s
OSDI 2012
Commit wait
average ε
average ε
14

15. Commit Wait and Replication
OSDI 2012
T
Start consensus Notify slaves
Acquired locks Release locks
Pick s Commit wait done
15
Achieve consensus

16. Commit Wait and 2-Phase Commit
TC
OSDI 2012
Acquired locks Release locks
TP1
Notify participants of s
Acquired locks Release locks
TP2
Acquired locks Release locks
Compute s for each Commit wait done
16
Start logging Done logging
Prepared
Compute overall s
Committed
Send s

17. Example
TC T2
TP
Remove X from
my friend list
Risky post P
s=8 s=15
Remove myself
from X’s friend list
sC=6
sP=8
s=8
Time <8
[X]
[me]
15
[P]
My friends
My posts
X’s friends
8
[]
[]
OSDI 2012 17

18. What Have We Covered?
• Lock-free read transactions across datacenters
• External consistency
• Timestamp assignment
• TrueTime
– Uncertainty in time can be waited out
OSDI 2012 18

19. What Haven’t We Covered?
• How to read at the present time
• Atomic schema changes
– Mostly non-blocking
– Commit in the future
• Non-blocking reads in the past
– At any sufficiently up-to-date replica
OSDI 2012 19

20. TrueTime Architecture
GPS
timemaster
GPS
timemaster
GPS
timemaster
Atomic-clock
timemaster
GPS
timemaster
GPS
timemaster
Client
Datacenter 1 Datacenter 2 … Datacenter n
Compute reference [earliest, latest] = now ± ε
OSDI 2012 20

21. TrueTime implementation
now = reference now + local-clock offset
ε = reference ε + worst-case local-clock drift
200 μs/sec
time
ε
0sec 30sec 60sec 90sec
+6ms
reference
uncertainty
OSDI 2012 21

22. What If a Clock Goes Rogue?
• Timestamp assignment would violate external
consistency
• Empirically unlikely based on 1 year of data
– Bad CPUs 6 times more likely than bad clocks
OSDI 2012 22

23. Network-Induced Uncertainty
10
8
6
4
Mar 29 Mar 30 Mar 31 Apr 1
OSDI 2012
Date
2
Epsilon (ms)
99.9
99
90
6
5
4
3
2
6AM 8AM 10AM 12PM
Date (April 13)
1
23

24. What’s in the Literature
• External consistency/linearizability
• Distributed databases
• Concurrency control
• Replication
• Time (NTP, Marzullo)
OSDI 2012 24

25. Future Work
• Improving TrueTime
– Lower ε < 1 ms
• Building out database features
– Finish implementing basic features
– Efficiently support rich query patterns
OSDI 2012 25

26. Conclusions
• Reify clock uncertainty in time APIs
– Known unknowns are better than unknown
unknowns
– Rethink algorithms to make use of uncertainty
• Stronger semantics are achievable
– Greater scale != weaker semantics
OSDI 2012 26

27. Thanks
• To the Spanner team and customers
• To our shepherd and reviewers
• To lots of Googlers for feedback
• To you for listening!
• Questions?
OSDI 2012 27